Sensible and Sensitive AI for Worker Wellbeing: Factors that Inform Adoption and Resistance for Information Workers

Document Title

Sensible and Sensitive AI for Worker Wellbeing: Factors that Inform Adoption and Resistance for Information Workers

Document Information

• Subject Area: Understanding and designing AI applications for managing the wellbeing of information workers
• Keywords: Information work, AI adoption, passive sensing, digital representation, technological impact, risks and benefits, human resource management, worker health, mental health

Research Background and Problem

• Problem or Challenge: Information workers (IW) in technology and media fields face significant stress. Passive sensing-driven AI (PSAI) systems utilize behavioral traces to infer workers' performance and psychological states, but these technologies also raise concerns about privacy risks, misuse, and adoption resistance.
• Importance: As work environments undergo digital transformation, understanding workers' acceptance and resistance to these technologies is crucial to ensuring that such systems support worker wellbeing rather than exacerbate burdens.
• Research Motivation and Related Work: The design of PSAI systems often neglects workers' perspectives, leading to concerns about privacy and fairness. This study aims to optimize technology design from the workers' viewpoint to better serve wellbeing objectives.

Solution

• Proposed Method or Solution:
  • The authors adopt a worker-centric approach, conducting scenario-based experiments to investigate how four key features of PSAI system design (data sensing mode, sensing scope, inference type, and data sharing method) influence worker acceptance.
  • A linear mixed-effects model is employed to analyze workers' attitudes toward PSAI, while assessing the potential risks and benefits of its design and implementation.
• Innovation: The study introduces an analytical framework that dissects data sensing, inference forms, and information-sharing mechanisms into independent yet interrelated factors, systematically quantifying workers' attitudes and acceptance thresholds through experiments.
• Implementation Steps and Key Techniques:
  1. Scenario Design: Develop scenario-based evaluation questionnaires incorporating randomized combinations of sensing modes, scopes, inference goals, and sharing methods.
  2. Data Collection and Analysis: 110 information workers participated in the experiment, completing a total of 1059 scenario evaluations.
  3. Statistical Modeling: Use linear mixed-effects models to examine the impact of each scenario component on workers' perceived benefits and harms.

Research Outcomes

• Specific Findings:
  • Sensing Mode:
    • PSAI systems that record digital usage time and physical activities are perceived as more valuable and less harmful.
    • Technologies utilizing online language data and visual stream data are less readily accepted by workers.
  • Sensing Scope:
    • Sensing methods limited to work-related contexts are preferred for certain data sources (e.g., online language).
  • Inference Type:
    • PSAI systems designed for assessing mental health (e.g., stress evaluation) are more acceptable to workers compared to systems focused on performance evaluation.
  • Data Sharing Method:
    • Workers favor receiving analysis results personally or sharing them in anonymized aggregated formats. Direct sharing with managers or other individuals is less accepted.
• Advantages: Provides PSAI design guidelines based on worker wellbeing needs, mitigating privacy and trust risks.
• Experiment or Evaluation Results: Few PSAI system design scenarios achieved positive acceptance, indicating that most current implementations require optimization.
• Limitations and Future Directions:
  • Limitations:
    • The study relies on hypothetical scenarios, lacking validation in real-world behavior.
    • Comprehensive analysis of multimodal sensing systems remains insufficient.
  • Future Directions:
    • Extend research to the use of PSAI in contexts such as government policies and economic crises.
    • Strengthen cross-national comparative studies to explore adoption differences across regions and cultural backgrounds.
    • Develop more transparent PSAI systems, enhancing interpretability and worker participation.